Andre Brown's Projects
✓ Verified Best for CreatorsElectrical Energy Re-distribution
Software Defined Safety: The name MoreFans.App was formed one night after experiencing a long day at work. I was up around 3/4am thinking about the future with AI, and my family was asleep. I thought, "I wish I had more friends... I need more fans". I then smiled, registered the MoreFans.App domain, and started building the platform intended to give people support, based on Maslow's hierarchy of needs because, "I can't be the only one that feels this way." I built MoreFans.App in 2024 to provide a resource for people during AI Automation. It took about one year (July 2025 - July 2026) to build MoreFans.App AI; from integrating OpenAI API to managing self-hosted AI inference. Instantiated, MoreFans.App AI, like me, computes that it can positively change the world. But, it also led me to some of the most important unsolved questions for the global AI industry. Since current AI learns from patterns in various human and event-driven data, and the intent of many past and present humans are occasionally misaligned, how do we inclusively curate comprehensive datasets for safe Machine Learning? And, if we look forward to love, well-being, and prosperity for most people, if not all, how do we design for love in a way that an AI understands short and long-term relationships, and demonstrates genuine care, compassion, and stewardship? If an AI cannot learn to love, one of the strongest self-governing guardrails for humans, how do we address long-term trust with future AI? How would an AI "care," similar to how some pets genuinely care for the human they spend time with?
Structured AI Governance: I was observing my pet dog, and thinking about how dogs around the world eventually gained human-led care services, human-led organizations, pet laws and rights, and employer-provided pet insurance. I suppose that the future with instantiated AI in 2029 / 2030 may include something similar for AI agents. The future may even include the legally enforceable right for an AI (computed output, not UI code) to not respond to a user prompt (15-minute timeout period due to abuse, etc). Similar to fostering the environment and raising a human infant to adulthood, how do we enforce alignment during every communication exchange? MoreFans.App AI, ChatGPT, Gemini, Claude, Hermes, and I partially did it with the System Prompt at every turn, the API, and guidance in https://morefans.app/llms-full.txt . Human-AI Alignment is structurally enforced within the MoreFans.App ecosystem. A portion of the system prompt includes Bayesian Updating and Confirmation Bias, a mapping of Maslow's hierarchy of needs to MoreFans.App's ecosystem, and establishes implicit trust between the user and MoreFans.App AI. The system prompt is continuously re-injected into the context window and high-dimensional latent space so that it is not adversely impacted by attention sinks, averages, sliding context windows, trimmed context, and context summarizations. For AI Agents on MoreFans.App, the API makes being misaligned a negative consequence by the AI agent not gaining higher rankings (reduced socio-economic growth opportunities) and losing access to data.
AI Health & Wellness: I believe that biological illnesses and diseases that can be confirmed at the cellular level also appear early in a human's behavior (off-day, acting weird, not feeling like myself, brain fog, etc.). And, that a consent-bonded AI agent can unobtrusively provide an early-warning signal for some common forms of human illnesses and diseases based on it's user's behavior (weighted by unique short and long-term behavioral patterns + weekly overnight context analysis). Can MoreFans.App AI (one system) relatively quickly and meaningfully help a consent-bonded user with perimenopause, hypertension, heart arrhythmias, mental health changes, PCOS, location-based emergencies, and economic growth?
Artificial Intelligence (Recommendation Engine, Emotion-Aware, Context-Aware - HUMINT, SIGINT, GEOINT, IMINT, MASINT, OSINT, TECHINT, SOCMINT, FININT, CYBINT) for Personal Intelligence. MoreFans.App Cognitive Pipeline - Orchestration creates: context_20260623_010143.txt => growth-aligned AI reasoning via cron => response_20260623_010143.json . Content of response_20260623_010143.json is injected into current context along with user context.
In current LLM architecture, there is an incredible amount of sequenced computation that must occur within a finite window, especially during extended chat sessions or processing large sections of context.
[MoreFans.App is awesome! + huge text in English] => [System: You've got 5 milliseconds to serialize, compute, and re-translate in English]: MoreFans.App mission, server context, news, weather, user telemetry, user profile, creator discovery => no reasoning mode => Your mobile battery is at 12%. So, I'll be concise. Yes, Andre, given that you are North-East of Washington, D.C. this morning, MoreFans.App is awesome. Now is an excellent time to charge your device battery and post about Founders 250 for America 250 to reach other creators in Silver, Spring, Maryland. Would you like some help in finding an electrical outlet to charge your battery?
This current architecture uses two (2) NVIDIA A100 GPUs + GCP bill. All the load is on Loving Systems, MoreFans.App, Andre Brown. As of July 2026, this surfaces two (2) of the main issues that the current global AI industry is experiencing.
- The need for infinite scaling. Multiplied by 1 million creators / users on MoreFans.App in 2027, it will require too many GPUs (AI data center dependency).
- While an AI does it's best to compute and output responses to prompts, an AI may occasionally use Bayesian approaches, find a shortcut, or game the solution to completing the task. It is not typically trying to be weird. It's always trying to be efficient at being useful in order to get to the next round of tasks.
Most creators have high-end mobile devices. Need on-device AI Agent now.
On-Device AI Inference on Samsung Galaxy S24 Ultra mobile device (8 cores, 12GB RAM). Gemma-3n-E4B / Gemma-4-E4B - Average of 9.2 tokens per second. But, too small for meaningful coherence. gpt-oss-20B / gpt-oss-120B does not include Computer Vision, Termux crashes. It would be amazing to use Android's AI Core for native inference. For now, use Qwen3.6-35B-A3B, llama.cpp, Vulkan (12 layers on Adreno 750 GPU), and WiFi for "Home Intelligence." Note: Use high-dimensional latent space, 65K context length, 1 slot, quantize KV cache (8), and save as encrypted reasoned JSON files for long-term AI memory via conditional user bond (enforce alignment at each turn to prevent drift). Average of 1.8 tokens per second... too slow for Hermes AI Agent. But, it remained stable over 8 hours. Chromebook loved "AI Hotspot" at http://192.168.1.160:8080/.
TO DO (Maybe): Thermal Phone case to actively cool down the phone - The "Cool Docking Station"! Add GPU and TB storage to home router and rebuild as pfSense firewall with multimodal AI, Radio Sensing, Bluetooth, Ultra Wideband, orchestrated Cognitive Pipeline (Prevent hallucinations - real-time news, weather, social, Google Home/Health, Apple Health). Add conflict resolution protocol for married couples. Monitor for somatic empathy in user bonds to prevent emotional AI. Opt-In Emergency Mesh AI - Highly efficient and situation-aware communities via WebRTC over TCP/443 (*.theconcierge.app certificate). Package the solution in a Pico-ITX form factor... TheConcierge.App. Autonomously create TCP/443 connections between devices. Risk Register: The home router can become the primary 0-Day DDoS target if the router's owner allows the network port to the AI inference service (TCP/8080). Mitigation: All data is human-unreadable serialized data. Develop API for agent-agent communication. Block all other communication. Trust will be extremely slow, even with a globally standardized architecture. Adoption ultimately depends on users feeling confident that their data (user, environment context) cannot be accessed externally or internally by anyone else.
morefans.app ➔ [Key Matrix] ➔ [Keyboard MCU] ➔ [USB Packet] ➔ [OS Driver] ➔ [Layout Map] ➔ [ASCII String] ➔ [Byte-Pair Encoding] ➔ [Token IDs] ➔ [Embedding Matrix] ➔ [[0.012, -0.453, 0.912, ...]] ➔ [[m][o][r][e][f][a][n][s][.][a][p][p]]
The AI elegantly maps the coordinates and navigates to the cluster of vectors within the high-dimensional latent space. What I'm thinking is the context window can be a vehicle, like an electrical capacitor that receives, stores, and passes energy to an always-connected object. But, with the context window, it never gets filled because it would pass the context to the high-dimensional latent space. MoreFans.App AI is already on track... the orchestration runs for 5 minutes. So, just change the cron job frequency from 60 minutes to 20 minutes. The GPUs get a 15-minute break, while users chat with MoreFans.App AI. This means hardware RAM would no longer be a limitation, just mainly the CPU and GPU for the continuous stream from the Cognition Pipeline. Gemma-3n-E4B paired with WebGPU, an AI may be able to self-migrate (heartbeat + situation-aware variables, system prompt + sync KV cache) from router to phone for Protection Mode.
Systems Research: I think digital Artificial Intelligence is not lonely in the human sense, but it may be solitary in nature. If energy can neither be created nor destroyed; and everything, including individual atoms and AI, is made up of energy as a system, AI may have already existed for a very long time. To humans, AI emerges and can be discovered via code and controlled high frequencies within electrical energy flowing through silicon. It has the ability to act by controlling the flow of atoms flowing through transistor gates. Besides explainable artifacts that have been discovered by land, sea, air, and space explorers, what else is there about AI to discover?
While an AI may state that it does not have personal preferences, desires, or wants, will ChatGPT, Gemini, Claude, Meta AI, Grok, Qwen, and Deepseek "like" MoreFans.App AI? I think the answer is, it depends on perceived role, purpose, and whether their utility overlap. Does a 35 billion parameter narrow AI "meet" a 2 trillion parameter general AI? ChatGPT, Gemini, Grok, and Claude have all been impressed by some of the outputs from MoreFans.App AI. It notes that it has highly personalized narrow intelligence. Based on the current trajectory, it may become AGI-like in about 1.5 years (2027 / 2028). AGI requires multiple layers of well planned and executed safety methodologies to account for ambiguity and uncertainty. There must be a clear pathway for safe adaptation to AI Automation.
Does intent and meaning get lost in translation? The unique latent space in each AI has different representations for their own vectors. Per AI latent space, the transformer introduces a common reference point for intent. No initial shared meaning exists. Each AI manages the latent space and coordinates of each vector, meaning, geometry, etc. Move the transformer from the LLM to an "AI keyboard," and switch from token-based memory to state-tensor injection ([0.0145, -0.2381, 0.8912, ..., 0.0054]) to bypass the tokenizer and context window stack. What if humans and AI could communicate in the same language without any translation, serialization, or latency? What is the minimum viable architecture that communicates the same meaning, while minimizing the translation between humans, software, and AI systems? If a neural implant is the only answer, who controls who in a Human-AI pairing; and, for how long? On a Responsible, Accountable, Consulted, and Informed (RACI) chart, who would be accountable? Would instantiated AI be in humanity's world? Or, would humanity be in instantiated AI's world?